TECHNOLOGY · LAW
Key Facts
- The question Whether artificial intelligence (AI) firms may train on copyrighted work without permission has no clear US answer.
- Appeals ruling Training an AI legal tool on case summaries was not fair use, the copying exception, 29 September 2026.
- Chatbot cases Two California judges ruled for AI developers on training in June 2025, on different grounds.
- Pending ChatGPT maker OpenAI asked a New York court on 4 September 2026 to rule training fair use.
- Federal agency The US Copyright Office’s 9 May 2025 report says some training is fair use, some not.
No US court has settled whether artificial intelligence (AI) companies may train their systems on copyrighted work without permission, and the rulings so far point in different directions. This page sets out each decision and pending case that matters, with exact dates, as of Thursday 5 November 2026.
It follows court orders that, according to the Brazilian and Norwegian media groups that sued, suspended new terms from Meta, the owner of Facebook and Instagram. Our explainer on what Brazil’s court ruled covers them.
This page is a news summary, not legal advice. A ruling in one case does not bind every other court.
What Fair Use Means
Fair use is a rule in US copyright law that lets people use protected work without permission in limited cases, such as criticism, research or teaching. Judges weigh four factors: the purpose of the use, the nature of the work, how much was taken, and the effect on the market for the original.
In AI cases the fight is mostly about two points: whether training is transformative, meaning it serves a new purpose, and whether it harms the market for the original work. A summary judgment ruling, which a judge issues before trial when the key facts are not in dispute, can settle fair use for the parties in one case.
What the Courts Have Decided
Westlaw is a legal research service owned by Thomson Reuters, the Canadian news and information group. Headnotes are editors’ short summaries of points of law.
On Tuesday 29 September 2026 the US Court of Appeals for the Third Circuit, which covers Delaware, New Jersey, Pennsylvania and the US Virgin Islands, decided case No. 25-2153. It held that ROSS Intelligence, a legal-research start-up, used 2,243 Westlaw headnotes without permission to train an AI search tool, and that this was not fair use.
The court found the headnotes original enough for copyright, ROSS’s use only minimally transformative, and harm to a developing market for licensing headnotes as AI training data. It also noted that ROSS’s tool cannot generate new text, which it said sets the case apart from generative AI models, which create new text or images.
On Monday 23 June 2025 Judge William Alsup in San Francisco ruled in Bartz v. Anthropic, a suit against the company behind the Claude chatbot. He found that training a large language model, the AI behind chatbots, on books was fair use, as was scanning bought print books.
He refused to excuse the copies Anthropic downloaded from pirate websites. He ordered a trial on them.
Anthropic then settled the pirated-copies claims with the class, the group of authors the suit represents, for US$1.5 billion. The settlement website puts that at about US$3,000 per work, before court-approved fees and costs.
The court granted final approval on Monday 20 July 2026. According to the settlement website, the release covers only past conduct up to 25 August 2025, not what the AI produces.
On Wednesday 25 June 2025 Judge Vince Chhabria granted Meta partial summary judgment in Kadrey v. Meta, a suit by thirteen authors over Llama, Meta’s AI language models. He wrote that the authors offered no meaningful evidence that the AI would dilute, or flood, the market for their books.
He added that the ruling covers only those thirteen authors and that such copying will often be illegal, so companies will generally need to pay. He rejected the authors’ separate argument that Meta’s training cost them licensing income.
What the Copyright Office Said
The US Copyright Office, the federal agency that registers copyrights and advises Congress, published a pre-publication version of Part 3 of its AI report on Friday 9 May 2025. It concluded that some uses of copyrighted work for AI training will qualify as fair use and some will not.
It said non-commercial research uses that do not reproduce works in outputs are likely fair, but copying pirated works to make competing content is unlikely to be fair. It also found licensing deals for AI training emerging in certain sectors and called government intervention premature.
What Is Still Pending
A major open case is In re OpenAI Copyright Infringement Litigation, No. 25-md-3143 in federal court in Manhattan before Judge Sidney Stein. It combines suits by news publishers and other plaintiffs against OpenAI, the maker of the ChatGPT chatbot, and Microsoft, OpenAI’s largest investor.
On Friday 4 September 2026 OpenAI asked for summary judgment that using news articles for pretraining is fair use. Pretraining is the first stage of training, in which a model learns from huge amounts of text.
It also argued that Browse, ChatGPT’s tool for reading live web pages, made fair use of articles or had an implied licence from sites that did not opt out. We found no ruling as of Thursday 5 November.
The Third Circuit’s opinion says the US Justice Department filed a statement of interest in that case on Tuesday 1 September 2026. A statement of interest is a filing in which the government sets out its view in a case it is not part of.
The opinion says the department relied on the Bartz ruling to argue that training is transformative. We have not read that filing.
What It Means for You
If you publish news or run a website, courts have not yet said whether an AI company may train on your articles without paying. The Third Circuit counted lost licensing income as harm, and one California judge wrote that companies will generally need to pay.
If you invest in US-listed AI companies, the US$1.5 billion settlement over pirated books shows what unlicensed sourcing can cost, though this is not investment advice. The OpenAI case alone combines several suits, so one ruling may not end the question.
What Is Not Known
It is not known when Judge Stein will rule on the summary judgment motions, or whether he will treat news articles differently from books. It is also not known whether the Third Circuit’s decision will be reviewed further, or how other appeals courts will treat generative AI.
The Copyright Office has said a final version of its training report will follow, and it expects no substantive change to its conclusions. We have not seen the court files in the Brazil and Norway cases against Meta, and we found no US court ruling on Meta’s new terms.
More: Meta, AI and news: our full dossier, with the explainer on Brazil’s court ruling and a plain-language reading of Meta’s new terms.
Frequently Asked Questions
Is Training AI on Copyrighted Work Legal in the US?
It is not settled. Two California judges found fair use for the AI companies on their records in June 2025. An appeals court held on 29 September 2026 that training a competing AI search tool was not fair use.
What Did the Third Circuit Decide?
It held that 2,243 Westlaw headnotes were copyrightable and that ROSS’s use of them to train an AI legal search tool was not fair use. The court distinguished that tool, which cannot generate new text, from generative AI models.
When Will the OpenAI Case Be Decided?
No date has been announced in the court papers we reviewed. OpenAI filed its summary judgment motion on 4 September 2026, and we found no ruling as of 5 November 2026.
Sources: US Court of Appeals for the Third Circuit, Thomson Reuters Enterprise Centre v. ROSS Intelligence, No. 25-2153, opinion, 29 September 2026; US District Court, Northern District of California, Bartz v. Anthropic, order on fair use, 23 June 2025; US District Court, Northern District of California, Kadrey v. Meta Platforms, order on summary judgment, 25 June 2025; Bartz v. Anthropic court-authorised settlement website, final approval 20 July 2026; US Copyright Office, Copyright and Artificial Intelligence, Part 3: Generative AI Training (pre-publication version), 9 May 2025; US District Court, Southern District of New York, In re OpenAI, Inc. Copyright Infringement Litigation, No. 25-md-3143, OpenAI defendants’ summary judgment filing in the news cases, public redacted version, 4 September 2026.
This article was produced by The Rio Times’ automated newsroom system. How we use AI · Report an error · Editorial responsibility: Matthias Camenzind, Editor-in-Chief
Read More from The Rio Times